Merge "11"
diff --git a/JWLLL/main_online.py b/JWLLL/main_online.py
new file mode 100644
index 0000000..0c2dd7b
--- /dev/null
+++ b/JWLLL/main_online.py
@@ -0,0 +1,1017 @@
+# main_online.py
+# 搜索推荐算法服务的主入口
+
+import json
+import numpy as np
+import difflib
+from flask import Flask, request, jsonify, Response
+import pymysql
+import jieba
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+import pypinyin
+from flask_cors import CORS
+import re
+import Levenshtein
+import os
+import logging
+
+# 设置日志
+logging.basicConfig(
+ level=logging.INFO,
+ format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
+)
+logger = logging.getLogger("allpt-search")
+
+# 导入Word2Vec辅助模块
+try:
+ from word2vec_helper import get_word2vec_helper, expand_query, get_similar_words
+ WORD2VEC_ENABLED = True
+ logger.info("Word2Vec模块已加载")
+except ImportError as e:
+ logger.warning(f"Word2Vec模块加载失败: {e}，将使用传统搜索")
+ WORD2VEC_ENABLED = False
+
+# 数据库配置
+DB_CONFIG = {
+ "host": "10.126.59.25",
+ "port": 3306,
+ "user": "root",
+ "password": "123456",
+ "database": "redbook",
+ "charset": "utf8mb4"
+}
+
+def get_db_conn():
+ return pymysql.connect(**DB_CONFIG)
+
+def get_pinyin(text):
+ # 返回字符串的全拼音（不带声调，全部小写），支持英文直接返回
+ if not text:
+ return ""
+ import re
+ # 如果全是英文，直接返回小写
+ if re.fullmatch(r'[a-zA-Z]+', text):
+ return text.lower()
+ return ''.join([p[0] for p in pypinyin.pinyin(text, style=pypinyin.NORMAL)])
+
+def get_pinyin_initials(text):
+ # 返回字符串的首字母拼音（全部小写），支持英文直接返回
+ if not text:
+ return ""
+ import re
+ if re.fullmatch(r'[a-zA-Z]+', text):
+ return text.lower()
+ return ''.join([p[0][0] for p in pypinyin.pinyin(text, style=pypinyin.NORMAL)])
+
+# 新增词语相似度计算函数
+def word_similarity(word1, word2):
+ """计算两个词的相似度，支持拼音匹配"""
+ # 直接匹配
+ if word1 == word2:
+ return 1.0
+
+ # 拼音匹配
+ if get_pinyin(word1) == get_pinyin(word2):
+ return 0.9
+
+ # 拼音首字母匹配
+ if get_pinyin_initials(word1) == get_pinyin_initials(word2):
+ return 0.7
+
+ # 字符串相似度
+ return difflib.SequenceMatcher(None, word1, word2).ratio()
+
+def semantic_title_similarity(query, title):
+ """计算查询词与标题的语义相似度"""
+ # 分词
+ query_words = list(jieba.cut(query))
+ title_words = list(jieba.cut(title))
+
+ if not query_words or not title_words:
+ return 0.0
+
+ # 计算每个查询词与标题词的最大相似度
+ max_similarities = []
+ key_matches = 0 # 关键词精确匹配数量
+
+ for q_word in query_words:
+ if len(q_word.strip()) <= 1: # 忽略单字，减少噪音
+ continue
+
+ word_sims = [word_similarity(q_word, t_word) for t_word in title_words]
+ if word_sims:
+ max_sim = max(word_sims)
+ max_similarities.append(max_sim)
+ if max_sim > 0.85: # 认为是关键词匹配
+ key_matches += 1
+
+ if not max_similarities:
+ return 0.0
+
+ # 计算平均相似度
+ avg_sim = sum(max_similarities) / len(max_similarities)
+
+ # 权重计算: 平均相似度占70%，关键词匹配率占30%
+ key_match_ratio = key_matches / len(query_words) if query_words else 0
+
+ # 标题中包含完整查询短语时给予额外加分
+ exact_bonus = 0.3 if query in title else 0
+
+ return 0.7 * avg_sim + 0.3 * key_match_ratio + exact_bonus
+
+# 添加语义关联词典，用于增强搜索能力
+def load_semantic_mappings():
+ """
+ 加载语义关联映射表，用于增强搜索语义理解
+ 返回包含语义映射关系的字典
+ """
+ # 初始化空字典，所有映射将从配置文件加载
+ mappings = {}
+
+ # 从配置文件加载映射
+ try:
+ config_path = os.path.join(os.path.dirname(__file__), "semantic_config.json")
+ if os.path.exists(config_path):
+ with open(config_path, 'r', encoding='utf-8') as f:
+ mappings = json.load(f)
+ logger.info(f"已从配置文件加载 {len(mappings)} 个语义映射")
+ else:
+ logger.warning(f"语义配置文件不存在: {config_path}")
+ except Exception as e:
+ logger.error(f"加载语义配置文件失败: {e}")
+
+ return mappings
+
+# 初始化语义映射
+SEMANTIC_MAPPINGS = load_semantic_mappings()
+
+def expand_search_keywords(keyword):
+ """
+ 扩展搜索关键词，增加语义关联词
+ """
+ expanded = [keyword]
+
+ # 分词处理
+ words = list(jieba.cut(keyword))
+ logger.info(f"关键词 '{keyword}' 分词结果: {words}") # 记录分词结果
+
+ # 分别对每个分词进行语义扩展
+ for word in words:
+ if word in SEMANTIC_MAPPINGS:
+ # 添加语义关联词
+ mapped_words = SEMANTIC_MAPPINGS[word]
+ expanded.extend(mapped_words)
+ logger.info(f"语义映射: '{word}' -> {mapped_words}")
+
+ # 移除所有特殊处理部分
+ # 不再对任何特定关键词如"越狱"进行特殊处理
+
+ # Word2Vec扩展 - 如果可用，对分词结果进行Word2Vec扩展
+ if WORD2VEC_ENABLED:
+ try:
+ # 使用单独的变量记录原始扩展结果，方便记录日志
+ original_expanded = set(expanded)
+
+ # 首先尝试对整个关键词进行扩展
+ w2v_expanded = set()
+ similar_words = get_similar_words(keyword, topn=3, min_similarity=0.6)
+ w2v_expanded.update(similar_words)
+
+ # 然后对较长的分词进行扩展
+ for word in words:
+ if len(word) > 1: # 忽略单字
+ similar_words = get_similar_words(word, topn=2, min_similarity=0.65)
+ w2v_expanded.update(similar_words)
+
+ # 合并结果
+ expanded.extend(w2v_expanded)
+
+ # 记录日志
+ if w2v_expanded:
+ logger.info(f"Word2Vec扩展: {keyword} -> {list(w2v_expanded)}")
+ except Exception as e:
+ # 出错时记录但不中断搜索流程
+ logger.error(f"Word2Vec扩展失败: {e}")
+ logger.info("将仅使用配置文件中的语义映射")
+
+ # 去重
+ return list(set(expanded))
+
+# 替换原有的calculate_keyword_relevance函数，采用更通用的相关性算法
+def calculate_keyword_relevance(keyword, item):
+ """计算搜索关键词与条目的相关性得分"""
+ title = item.get('title', '')
+ description = item.get('description', '') or ''
+ tags = item.get('tags', '') or ''
+ category = item.get('category', '') or '' # 添加category字段
+
+ # 初始化得分
+ score = 0
+
+ # 1. 精确匹配（最高优先级）
+ if keyword.lower() == title.lower():
+ return 15.0 # 完全匹配给予最高分
+
+ # 2. 标题中精确词匹配
+ title_words = re.findall(r'\b\w+\b', title.lower())
+ if keyword.lower() in title_words:
+ score += 10.0 # 作为独立词完全匹配
+
+ # 3. 标题包含关键词（部分匹配）
+ elif keyword.lower() in title.lower():
+ # 计算关键词所占标题比例
+ match_ratio = len(keyword) / len(title)
+ if match_ratio > 0.5: # 关键词占标题很大比例
+ score += 8.0
+ else:
+ score += 5.0
+
+ # 4. 标题分词匹配
+ keyword_words = list(jieba.cut(keyword))
+ title_jieba_words = list(jieba.cut(title))
+
+ matched_words = 0
+ for k_word in keyword_words:
+ if len(k_word) > 1: # 忽略单字
+ if k_word in title_jieba_words:
+ matched_words += 1
+ else:
+ # 拼音匹配
+ k_pinyin = get_pinyin(k_word)
+ for t_word in title_jieba_words:
+ if get_pinyin(t_word) == k_pinyin:
+ matched_words += 0.8
+ break
+
+ if len(keyword_words) > 0:
+ word_match_ratio = matched_words / len(keyword_words)
+ score += 3.0 * word_match_ratio
+
+ # 5. 拼音相似度
+ keyword_pinyin = get_pinyin(keyword)
+ title_pinyin = get_pinyin(title)
+
+ if keyword_pinyin == title_pinyin:
+ score += 3.5
+ elif keyword_pinyin in title_pinyin:
+ # 计算拼音在标题中的位置影响
+ pos = title_pinyin.find(keyword_pinyin)
+ if pos == 0: # 出现在开头
+ score += 3.0
+ else:
+ score += 2.0
+
+ # 6. 编辑距离相似度
+ try:
+ edit_distance = Levenshtein.distance(keyword.lower(), title.lower())
+ max_len = max(len(keyword), len(title))
+ if max_len > 0:
+ similarity = 1 - (edit_distance / max_len)
+ if similarity > 0.7:
+ score += 1.5 * similarity
+ except:
+ similarity = difflib.SequenceMatcher(None, keyword.lower(), title.lower()).ratio()
+ if similarity > 0.7:
+ score += 1.5 * similarity
+
+ # 7. 中文字符重叠检测 - 修改为仅当重叠2个以上汉字或占比超过40%时才计分
+ if re.search(r'[\u4e00-\u9fff]', keyword) and re.search(r'[\u4e00-\u9fff]', title):
+ cn_chars_keyword = set(re.findall(r'[\u4e00-\u9fff]', keyword))
+ cn_chars_title = set(re.findall(r'[\u4e00-\u9fff]', title))
+
+ # 计算重叠的汉字集合
+ overlapped_chars = cn_chars_keyword & cn_chars_title
+
+ # 仅当重叠汉字数量大于1且占比超过阈值时才计分
+ if len(overlapped_chars) > 1 and len(cn_chars_keyword) > 0:
+ overlap_ratio = len(overlapped_chars) / len(cn_chars_keyword)
+ # 增加重叠比例的阈值要求，防止单个汉字导致的误匹配
+ if overlap_ratio >= 0.4 or len(overlapped_chars) >= 3:
+ score += 2.0 * overlap_ratio
+ # 对于非常低的重叠度，不加分，避免无关内容干扰
+
+ # 记录日志，帮助调试特定案例
+ if keyword == "明日方舟" and "白日梦想家" in title:
+ logger.info(f"'明日方舟'与'{title}'的汉字重叠: {overlapped_chars}, 重叠比例: {len(overlapped_chars)/len(cn_chars_keyword) if cn_chars_keyword else 0}")
+
+ # 8. 序列资源检测（如"功夫熊猫2"是"功夫熊猫"的系列）
+ base_title_match = re.match(r'(.*?)([0-9]+|[一二三四五六七八九十]|：|\:|\s+[0-9]+)', title)
+ if base_title_match:
+ base_title = base_title_match.group(1).strip()
+ if keyword.lower() == base_title.lower():
+ score += 2.0
+
+ # 9. 标签和描述匹配（增加权重）
+ if tags:
+ tags_list = tags.split(',')
+ if keyword in tags_list:
+ score += 1.5 # 提高标签匹配的权重
+ elif any(keyword.lower() in tag.lower() for tag in tags_list):
+ score += 1.0 # 提高部分匹配的权重
+
+ # 描述匹配增强
+ if keyword.lower() in description.lower():
+ score += 1.5 # 提高描述匹配的权重
+
+ # 检查关键词在描述中的位置和上下文
+ pos = description.lower().find(keyword.lower())
+ if pos >= 0 and pos < len(description) / 3:
+ # 关键词出现在描述前1/3部分，可能更重要
+ score += 0.5
+
+ # 考虑分词匹配描述
+ keyword_words = list(jieba.cut(keyword))
+ description_words = list(jieba.cut(description))
+ matched_desc_words = 0
+ for k_word in keyword_words:
+ if len(k_word) > 1 and k_word in description_words:
+ matched_desc_words += 1
+
+ if len(keyword_words) > 0:
+ desc_match_ratio = matched_desc_words / len(keyword_words)
+ score += 1.0 * desc_match_ratio
+
+ # 分类匹配
+ if keyword.lower() in category.lower():
+ score += 1.0
+
+ # 添加语义关联匹配得分
+ # 扩展关键词进行匹配
+ expanded_keywords = expand_search_keywords(keyword)
+
+ # 检测标题是否包含语义相关词
+ for exp_keyword in expanded_keywords:
+ if exp_keyword != keyword and exp_keyword in title: # 避免重复计算原关键词
+ # 根据关联词的匹配类型给予不同分数
+ if exp_keyword in ["国宝", "熊猫"] and "功夫熊猫" in title:
+ score += 3.0 # 高度相关的语义映射
+ elif exp_keyword in title:
+ score += 1.5 # 一般语义关联
+
+ # 对于特殊组合查询，额外加分
+ if ("国宝" in keyword or "熊猫" in keyword) and "电影" in keyword and "功夫熊猫" in title:
+ score += 4.0 # 对"国宝电影"、"熊猫电影"搜"功夫熊猫"特别加分
+
+ return score
+
+# 创建Flask应用
+app = Flask(__name__)
+CORS(app) # 允许所有跨域请求
+
+# 添加init_word2vec函数
+def init_word2vec():
+ """初始化Word2Vec模型"""
+ try:
+ helper = get_word2vec_helper()
+ if helper.initialized:
+ logger.info(f"Word2Vec模型已成功加载，词汇量: {len(helper.model.index_to_key)}, 向量维度: {helper.model.vector_size}")
+ else:
+ if helper.load_model():
+ logger.info(f"Word2Vec模型加载成功，词汇量: {len(helper.model.index_to_key)}, 向量维度: {helper.model.vector_size}")
+ else:
+ logger.error("Word2Vec模型加载失败")
+ except Exception as e:
+ logger.error(f"初始化Word2Vec出错: {e}")
+
+# 新的初始化方式:
+def initialize_app():
+ """应用初始化函数，替代before_first_request装饰器"""
+ # 修正：使用正确的函数名
+ # 原代码: init_semantic_mapping()
+ # 修正为使用已定义的函数名
+ global SEMANTIC_MAPPINGS
+ SEMANTIC_MAPPINGS = load_semantic_mappings() # 更新全局语义映射变量
+
+ if WORD2VEC_ENABLED:
+ init_word2vec() # 现在这个函数已经定义了
+
+# 在启动应用之前调用初始化函数
+initialize_app()
+
+# 搜索功能的API
+@app.route('/search', methods=['POST'])
+def search():
+ """
+ 搜索功能API
+ 请求格式：{
+ "keyword": "关键词",
+ "sort_by": "downloads" | "downloads_asc" | "newest" | "oldest" | "similarity" | "title_asc" | "title_desc",
+ "category": "可选，分类名",
+ "search_mode": "title" | "title_desc" | "tags" | "all" # 可选，默认"title",
+ "tags": ["标签1", "标签2"] # 可选，支持传递多个标签
+ }
+ """
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+
+ data = request.get_json()
+ keyword = data.get("keyword", "").strip()
+ sort_by = data.get("sort_by", "similarity") # 默认按相似度排序
+ category = data.get("category", None)
+ search_mode = data.get("search_mode", "title")
+ tags = data.get("tags", None) # 支持传递多个标签
+
+ # 校验参数 - 不管什么模式都要求关键词
+ if not (1 <= len(keyword) <= 20):
+ return jsonify({"error": "请输入1-20个字符"}), 400
+
+ # 第一阶段：数据库查询获取候选集
+ results = []
+ conn = get_db_conn()
+ try:
+ with conn.cursor(pymysql.cursors.DictCursor) as cursor:
+ # 首先尝试查询完全匹配的结果
+ exact_query = f"""
+ SELECT id, title, topic_id, heat, created_at, content
+ FROM posts
+ WHERE title = %s
+ """
+ cursor.execute(exact_query, (keyword,))
+ exact_matches = cursor.fetchall() or [] # 确保返回列表而非元组
+
+ # 扩展关键词，增加语义关联词
+ expanded_keywords = expand_search_keywords(keyword)
+ logger.info(f"扩展后的关键词: {expanded_keywords}") # 调试信息
+
+ # 构建查询条件
+ conditions = []
+ params = []
+
+ # 标题匹配 - 所有搜索模式都匹配title
+ conditions.append("title LIKE %s")
+ params.append(f"%{keyword}%")
+
+ # 为扩展关键词添加标题匹配条件
+ for exp_keyword in expanded_keywords:
+ if exp_keyword != keyword: # 避免重复原关键词
+ conditions.append("title LIKE %s")
+ params.append(f"%{exp_keyword}%")
+
+ # 描述匹配
+ if search_mode in ["title_desc", "all"]:
+ # 原始关键词匹配描述
+ conditions.append("content LIKE %s")
+ params.append(f"%{keyword}%")
+
+ # 扩展关键词匹配描述
+ for exp_keyword in expanded_keywords:
+ if exp_keyword != keyword:
+ conditions.append("content LIKE %s")
+ params.append(f"%{exp_keyword}%")
+
+ # 标签匹配
+ # 暂不处理，后续join实现
+
+ # 分类匹配 - 仅在all模式下
+ if search_mode == "all":
+ # 原始关键词匹配分类
+ conditions.append("topic_id LIKE %s")
+ params.append(f"%{keyword}%")
+
+ # 扩展关键词匹配分类
+ for exp_keyword in expanded_keywords:
+ if exp_keyword != keyword:
+ conditions.append("topic_id LIKE %s")
+ params.append(f"%{exp_keyword}%")
+
+ # 构建SQL查询
+ if conditions:
+ where_clause = " OR ".join(conditions)
+ logger.info(f"搜索条件: {where_clause}")
+ logger.info(f"参数列表: {params}")
+
+ if category:
+ where_clause = f"({where_clause}) AND topic_id=%s"
+ params.append(category)
+
+ sql = f"""
+ SELECT p.id, p.title, tp.name as category, p.heat, p.created_at, p.content,
+ GROUP_CONCAT(t.name) as tags
+ FROM posts p
+ LEFT JOIN post_tags pt ON p.id = pt.post_id
+ LEFT JOIN tags t ON pt.tag_id = t.id
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ WHERE {where_clause}
+ GROUP BY p.id
+ LIMIT 500
+ """
+
+ cursor.execute(sql, params)
+ expanded_results = cursor.fetchall()
+ logger.info(f"数据库返回记录数: {len(expanded_results) if expanded_results else 0}")
+ else:
+ expanded_results = []
+
+ # 如果扩展查询和精确匹配都没有结果，获取全部记录进行相关性计算
+ if not expanded_results and not exact_matches:
+ sql = "SELECT p.id, p.title, tp.name as category, p.heat, p.created_at, p.content, GROUP_CONCAT(t.name) as tags FROM posts p LEFT JOIN post_tags pt ON p.id = pt.post_id LEFT JOIN tags t ON pt.tag_id = t.id LEFT JOIN topics tp ON p.topic_id = tp.id"
+ if category:
+ sql += " WHERE p.topic_id=%s"
+ category_params = [category]
+ cursor.execute(sql + " GROUP BY p.id", category_params)
+ else:
+ cursor.execute(sql + " GROUP BY p.id")
+
+ all_results = cursor.fetchall() or [] # 确保返回列表
+ else:
+ if isinstance(exact_matches, tuple):
+ exact_matches = list(exact_matches)
+ if isinstance(expanded_results, tuple):
+ expanded_results = list(expanded_results)
+ all_results = expanded_results + exact_matches
+
+ # 对所有结果使用相关性计算规则
+ scored_results = []
+ for item in all_results:
+ # 计算相关性得分
+ relevance_score = calculate_keyword_relevance(keyword, item)
+
+ # 降低相关性阈值，确保更多结果被保留 (从0.5改为0.1)
+ if relevance_score > 0.1:
+ item['relevance_score'] = relevance_score
+ scored_results.append(item)
+ logger.info(f"匹配项: {item['title']}, 相关性得分: {relevance_score}")
+
+ # 按相关性得分排序
+ scored_results.sort(key=lambda x: x.get('relevance_score', 0), reverse=True)
+
+ # 确保精确匹配的结果置顶
+ if exact_matches:
+ for exact_match in exact_matches:
+ exact_match['relevance_score'] = 20.0 # 超高分确保置顶
+
+ # 移除scored_results中已经存在于exact_matches的项
+ exact_ids = {item['id'] for item in exact_matches}
+ scored_results = [item for item in scored_results if item['id'] not in exact_ids]
+
+ # 合并两个结果集
+ results = exact_matches + scored_results
+ else:
+ results = scored_results
+
+ # 限制返回结果数量
+ results = results[:50]
+
+ except Exception as e:
+ logger.error(f"搜索出错: {e}")
+ import traceback
+ traceback.print_exc()
+ return jsonify({"error": "搜索系统异常，请稍后再试"}), 500
+ finally:
+ conn.close()
+
+ # 第二阶段：根据指定方式排序
+ if results:
+ if sort_by == "similarity" or not sort_by:
+ # 保持按相关性得分排序，已经排好了
+ pass
+ elif sort_by == "downloads":
+ results.sort(key=lambda x: x.get("download_count", 0), reverse=True)
+ elif sort_by == "downloads_asc":
+ results.sort(key=lambda x: x.get("download_count", 0))
+ elif sort_by == "newest":
+ results.sort(key=lambda x: x.get("create_time", ""), reverse=True)
+ elif sort_by == "oldest":
+ results.sort(key=lambda x: x.get("create_time", ""))
+ elif sort_by == "title_asc":
+ results.sort(key=lambda x: x.get("title", ""))
+ elif sort_by == "title_desc":
+ results.sort(key=lambda x: x.get("title", ""), reverse=True)
+
+ # 最终处理：清理不需要返回的字段
+ for item in results:
+ item.pop("description", None)
+ item.pop("tags", None)
+ item.pop("relevance_score", None)
+
+ return Response(json.dumps({"results": results}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+# 推荐功能的API
+@app.route('/recommend_tags', methods=['POST'])
+def recommend_tags():
+ """
+ 推荐功能API
+ 请求格式：{
+ "user_id": "user1",
+ "tags": ["标签1", "标签2"] # 可为空
+ }
+ """
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+
+ data = request.get_json()
+ user_id = data.get("user_id")
+ tags = set(data.get("tags", []))
+
+ # 查询用户已保存的兴趣标签
+ user_tags = set()
+ if user_id:
+ conn = get_db_conn()
+ try:
+ with conn.cursor() as cursor:
+ cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
+ user_tags = set(row[0] for row in cursor.fetchall())
+ finally:
+ conn.close()
+
+ # 合并前端传递的tags和用户兴趣标签
+ all_tags = list(tags | user_tags)
+
+ if not all_tags:
+ return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200
+
+ conn = get_db_conn()
+ try:
+ with conn.cursor(pymysql.cursors.DictCursor) as cursor:
+ # 优先用tags字段匹配
+ # 先查找所有tag_id
+ tag_ids = []
+ for tag in all_tags:
+ cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
+ row = cursor.fetchone()
+ if row:
+ tag_ids.append(row['id'])
+ if not tag_ids:
+ return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200
+ tag_placeholders = ','.join(['%s'] * len(tag_ids))
+ sql = f"""
+ SELECT p.id, p.title, tp.name as category, p.heat,
+ GROUP_CONCAT(tg.name) as tags
+ FROM posts p
+ LEFT JOIN post_tags pt ON p.id = pt.post_id
+ LEFT JOIN tags tg ON pt.tag_id = tg.id
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ WHERE pt.tag_id IN ({tag_placeholders})
+ GROUP BY p.id
+ LIMIT 50
+ """
+ cursor.execute(sql, tuple(tag_ids))
+ results = cursor.fetchall()
+ # 若无结果，回退title/content模糊匹配
+ if not results:
+ or_conditions = []
+ params = []
+ for tag in all_tags:
+ or_conditions.append("p.title LIKE %s OR p.content LIKE %s")
+ params.extend(['%' + tag + '%', '%' + tag + '%'])
+ where_clause = ' OR '.join(or_conditions)
+ sql = f"""
+ SELECT p.id, p.title, tp.name as category, p.heat,
+ GROUP_CONCAT(tg.name) as tags
+ FROM posts p
+ LEFT JOIN post_tags pt ON p.id = pt.post_id
+ LEFT JOIN tags tg ON pt.tag_id = tg.id
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ WHERE {where_clause}
+ GROUP BY p.id
+ LIMIT 50
+ """
+ cursor.execute(sql, tuple(params))
+ results = cursor.fetchall()
+ finally:
+ conn.close()
+
+ if not results:
+ return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200
+
+ return Response(json.dumps({"recommendations": results}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+# 用户兴趣标签管理API（可选）
+@app.route('/tags', methods=['POST', 'GET', 'DELETE'])
+def user_tags():
+ """
+ POST: 添加用户兴趣标签
+ GET: 查询用户兴趣标签
+ DELETE: 删除用户兴趣标签
+ """
+ if request.method == 'POST':
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+ data = request.get_json()
+ user_id = data.get("user_id")
+ tags = data.get("tags", [])
+
+ if not user_id:
+ return jsonify({"error": "用户ID不能为空"}), 400
+
+ # 确保标签列表格式正确
+ if isinstance(tags, str):
+ tags = [tag.strip() for tag in tags.split(',') if tag.strip()]
+
+ if not tags:
+ return jsonify({"error": "标签不能为空"}), 400
+
+ conn = get_db_conn()
+ try:
+ with conn.cursor() as cursor:
+ # 添加用户标签
+ for tag in tags:
+ # 先查找tag_id
+ cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
+ tag_row = cursor.fetchone()
+ if tag_row:
+ tag_id = tag_row[0]
+ cursor.execute("REPLACE INTO user_tags (user_id, tag_id) VALUES (%s, %s)", (user_id, tag_id))
+ conn.commit()
+ # 返回更新后的标签列表
+ cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
+ updated_tags = [row[0] for row in cursor.fetchall()]
+ finally:
+ conn.close()
+ return Response(json.dumps({"msg": "添加成功", "tags": updated_tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ elif request.method == 'DELETE':
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+ data = request.get_json()
+ user_id = data.get("user_id")
+ tags = data.get("tags", [])
+ if not user_id:
+ return jsonify({"error": "用户ID不能为空"}), 400
+ if not tags:
+ return jsonify({"error": "标签不能为空"}), 400
+
+ conn = get_db_conn()
+ try:
+ with conn.cursor() as cursor:
+ for tag in tags:
+ cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
+ tag_row = cursor.fetchone()
+ if tag_row:
+ tag_id = tag_row[0]
+ cursor.execute("DELETE FROM user_tags WHERE user_id=%s AND tag_id=%s", (user_id, tag_id))
+ conn.commit()
+ cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
+ remaining_tags = [row[0] for row in cursor.fetchall()]
+ finally:
+ conn.close()
+ return Response(json.dumps({"msg": "删除成功", "tags": remaining_tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ else: # GET 请求
+ user_id = request.args.get("user_id")
+ if not user_id:
+ return jsonify({"error": "用户ID不能为空"}), 400
+ conn = get_db_conn()
+ try:
+ with conn.cursor() as cursor:
+ cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
+ tags = [row[0] for row in cursor.fetchall()]
+ finally:
+ conn.close()
+ return Response(json.dumps({"tags": tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+# 添加/user_tags路由作为/tags的别名
+@app.route('/user_tags', methods=['POST', 'GET', 'DELETE'])
+def user_tags_alias():
+ """
+ /user_tags路由 - 作为/tags路由的别名
+ POST: 添加用户兴趣标签
+ GET: 查询用户兴趣标签
+ DELETE: 删除用户兴趣标签
+ """
+ return user_tags()
+
+# 基于用户的协同过滤推荐API
+@app.route('/user_based_recommend', methods=['POST'])
+def user_based_recommend():
+ """
+ 基于用户的协同过滤推荐API
+ 请求格式：{
+ "user_id": "user1",
+ "top_n": 5
+ }
+ """
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+
+ data = request.get_json()
+ user_id = data.get("user_id")
+ top_n = int(data.get("top_n", 5))
+
+ if not user_id:
+ return jsonify({"error": "用户ID不能为空"}), 400
+
+ conn = get_db_conn()
+ try:
+ with conn.cursor(pymysql.cursors.DictCursor) as cursor:
+ # 1. 检查用户是否存在下载记录（收藏或浏览）
+ cursor.execute("""
+ SELECT COUNT(*) as count
+ FROM behaviors
+ WHERE user_id = %s AND type IN ('favorite', 'view')
+ """, (user_id,))
+ result = cursor.fetchone()
+ user_download_count = result['count'] if result else 0
+
+ logger.info(f"用户 {user_id} 下载记录数: {user_download_count}")
+
+ # 如果用户没有足够的行为数据，返回基于热度的推荐
+ if user_download_count < 3:
+ logger.info(f"用户 {user_id} 下载记录不足，返回热门推荐")
+ cursor.execute("""
+ SELECT p.id, p.title, tp.name as category, p.heat
+ FROM posts p
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ ORDER BY p.heat DESC
+ LIMIT %s
+ """, (top_n,))
+ popular_seeds = cursor.fetchall()
+ return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+ # 2. 获取用户已下载（收藏/浏览）的帖子
+ cursor.execute("""
+ SELECT post_id
+ FROM behaviors
+ WHERE user_id = %s AND type IN ('favorite', 'view')
+ """, (user_id,))
+ user_seeds = set(row['post_id'] for row in cursor.fetchall())
+ logger.info(f"用户 {user_id} 已下载种子: {user_seeds}")
+
+ # 3. 获取所有用户-帖子下载（收藏/浏览）矩阵
+ cursor.execute("""
+ SELECT user_id, post_id
+ FROM behaviors
+ WHERE created_at > DATE_SUB(NOW(), INTERVAL 3 MONTH)
+ AND user_id <> %s AND type IN ('favorite', 'view')
+ """, (user_id,))
+ download_records = cursor.fetchall()
+
+ if not download_records:
+ logger.info(f"没有其他用户的下载记录，返回热门推荐")
+ cursor.execute("""
+ SELECT p.id, p.title, tp.name as category, p.heat
+ FROM posts p
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ ORDER BY p.heat DESC
+ LIMIT %s
+ """, (top_n,))
+ popular_seeds = cursor.fetchall()
+ return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+ # 构建用户-物品矩阵
+ user_item_matrix = {}
+ for record in download_records:
+ uid = record['user_id']
+ sid = record['post_id']
+ if uid not in user_item_matrix:
+ user_item_matrix[uid] = set()
+ user_item_matrix[uid].add(sid)
+
+ # 4. 计算用户相似度
+ similar_users = []
+ for other_id, other_seeds in user_item_matrix.items():
+ if other_id == user_id:
+ continue
+ intersection = len(user_seeds.intersection(other_seeds))
+ union = len(user_seeds.union(other_seeds))
+ if union > 0 and intersection > 0:
+ similarity = intersection / union
+ similar_users.append((other_id, similarity, other_seeds))
+ logger.info(f"找到 {len(similar_users)} 个相似用户")
+ similar_users.sort(key=lambda x: x[1], reverse=True)
+ similar_users = similar_users[:5]
+ # 5. 基于相似用户推荐帖子
+ candidate_seeds = {}
+ for similar_user, similarity, seeds in similar_users:
+ logger.info(f"相似用户 {similar_user}, 相似度 {similarity}")
+ for post_id in seeds:
+ if post_id not in user_seeds:
+ if post_id not in candidate_seeds:
+ candidate_seeds[post_id] = 0
+ candidate_seeds[post_id] += similarity
+ if not candidate_seeds:
+ logger.info(f"没有找到候选种子，返回热门推荐")
+ cursor.execute("""
+ SELECT p.id, p.title, tp.name as category, p.heat
+ FROM posts p
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ ORDER BY p.heat DESC
+ LIMIT %s
+ """, (top_n,))
+ popular_seeds = cursor.fetchall()
+ return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ # 6. 获取推荐帖子的详细信息
+ recommended_seeds = sorted(candidate_seeds.items(), key=lambda x: x[1], reverse=True)[:top_n]
+ post_ids = [post_id for post_id, _ in recommended_seeds]
+ format_strings = ','.join(['%s'] * len(post_ids))
+ cursor.execute(f"""
+ SELECT p.id, p.title, tp.name as category, p.heat
+ FROM posts p
+ LEFT JOIN topics tp ON p.topic_id = tp.id
+ WHERE p.id IN ({format_strings})
+ """, tuple(post_ids))
+ result_seeds = cursor.fetchall()
+ seed_score_map = {post_id: score for post_id, score in recommended_seeds}
+ result_seeds.sort(key=lambda x: seed_score_map.get(x['id'], 0), reverse=True)
+ logger.info(f"返回 {len(result_seeds)} 个基于协同过滤的推荐")
+ return Response(json.dumps({"recommendations": result_seeds, "type": "collaborative"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ except Exception as e:
+ logger.error(f"推荐系统错误: {e}")
+ import traceback
+ traceback.print_exc()
+ return Response(json.dumps({"error": "推荐系统异常，请稍后再试", "details": str(e)}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ finally:
+ conn.close()
+@app.route('/word2vec_status', methods=['GET'])
+def word2vec_status():
+ """
+ 检查Word2Vec模型状态
+ 返回模型是否加载、词汇量等信息
+ """
+ if not WORD2VEC_ENABLED:
+ return Response(json.dumps({
+ "enabled": False,
+ "message": "Word2Vec功能未启用"
+ }, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ try:
+ helper = get_word2vec_helper()
+ status = {
+ "enabled": WORD2VEC_ENABLED,
+ "initialized": helper.initialized,
+ "vocab_size": len(helper.model.index_to_key) if helper.model else 0,
+ "vector_size": helper.model.vector_size if helper.model else 0
+ }
+
+ # 测试几个常用词的相似词，展示模型效果
+ test_results = {}
+ test_words = ["电影", "动作", "科幻", "动漫", "游戏"]
+ for word in test_words:
+ similar_words = helper.get_similar_words(word, topn=5)
+ test_results[word] = similar_words
+
+ status["test_results"] = test_results
+ return Response(json.dumps(status, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ except Exception as e:
+ return Response(json.dumps({
+ "enabled": WORD2VEC_ENABLED,
+ "initialized": False,
+ "error": str(e)
+ }, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+
+# 添加一个临时诊断端点
+@app.route('/debug_search', methods=['POST'])
+def debug_search():
+ """临时的调试端点，用于检查数据库中的记录"""
+ if request.content_type != 'application/json':
+ return jsonify({"error": "Content-Type must be application/json"}), 415
+
+ data = request.get_json()
+ keyword = data.get("keyword", "").strip()
+
+ conn = get_db_conn()
+ try:
+ with conn.cursor(pymysql.cursors.DictCursor) as cursor:
+ # 尝试查询包含特定词的所有记录
+ queries = [
+ ("标题中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE title LIKE '%{keyword}%' LIMIT 10"),
+ ("描述中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE description LIKE '%{keyword}%' LIMIT 10"),
+ ("标签中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE FIND_IN_SET('{keyword}', tags) LIMIT 10"),
+ ("肖申克的救赎", "SELECT seed_id, title, description, tags FROM pt_seed WHERE title = '肖申克的救赎'")
+ ]
+
+ results = {}
+ for query_name, query in queries:
+ cursor.execute(query)
+ results[query_name] = cursor.fetchall()
+
+ return Response(json.dumps(results, ensure_ascii=False), mimetype='application/json; charset=utf-8')
+ finally:
+ conn.close()
+
+"""
+接口本地测试方法（可直接运行main_online.py后用curl或Postman测试）：
+
+1. 搜索接口
+curl -X POST http://127.0.0.1:5000/search -H "Content-Type: application/json" -d '{"keyword":"电影","sort_by":"downloads"}'
+
+2. 标签推荐接口
+curl -X POST http://127.0.0.1:5000/recommend_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'
+
+3. 用户兴趣标签管理（添加标签）
+curl -X POST http://127.0.0.1:5000/user_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'
+
+4. 用户兴趣标签管理（查询标签）
+curl "http://127.0.0.1:5000/user_tags?user_id=1"
+
+5. 用户兴趣标签管理（删除标签）
+curl -X DELETE http://127.0.0.1:5000/user_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'
+
+6. 协同过滤推荐
+curl -X POST http://127.0.0.1:5000/user_based_recommend -H "Content-Type: application/json" -d '{"user_id":"user1","top_n":3}'
+
+7. Word2Vec状态检查
+curl "http://127.0.0.1:5000/word2vec_status"
+
+8. 调试接口（临时）
+curl -X POST http://127.0.0.1:5000/debug_search -H "Content-Type: application/json" -d '{"keyword":"电影"}'
+
+所有接口均可用Postman按上述参数测试。
+"""
+
+if __name__ == "__main__":
+ try:
+ logger.info("搜索推荐服务启动中...")
+ app.run(host="0.0.0.0", port=5000)
+ except Exception as e:
+ logger.error(f"启动异常: {e}")
+ import traceback
+ traceback.print_exc()
diff --git a/JWLLL/semantic_config.json b/JWLLL/semantic_config.json
new file mode 100644
index 0000000..9f54454
--- /dev/null
+++ b/JWLLL/semantic_config.json
@@ -0,0 +1,73 @@
+{
+ "国宝": ["熊猫", "大熊猫", "功夫熊猫", "四川", "成都", "保护动物"],
+ "熊猫": ["国宝", "大熊猫", "功夫熊猫", "竹子", "四川", "黑白"],
+ "功夫": ["武术", "格斗", "武打", "功夫熊猫", "李小龙", "成龙", "太极", "截拳道", "中国功夫"],
+
+ "梦": ["梦想", "梦境", "白日梦", "白日梦想家", "潜意识", "睡眠", "做梦"],
+ "白日梦": ["梦想", "幻想", "白日梦想家", "想象", "憧憬"],
+
+ "魔戒": ["指环王", "魔戒再现", "中土世界", "霍比特人", "精灵", "魔法", "奇幻"],
+ "指环": ["魔戒", "指环王", "戒指", "魔戒再现", "首饰"],
+ "中土世界": ["魔戒", "指环王", "霍比特人", "精灵", "矮人", "奇幻"],
+
+ "漫威": ["复仇者", "钢铁侠", "蜘蛛侠", "美国队长", "雷神", "绿巨人", "黑寡妇", "惊奇队长", "超级英雄", "漫画"],
+ "钢铁侠": ["托尼斯塔克", "钢铁战衣", "贾维斯", "复仇者", "漫威", "超级英雄"],
+ "蜘蛛侠": ["彼得帕克", "蜘蛛", "纽约", "漫威", "超级英雄", "蜘蛛感应"],
+
+ "DC": ["蝙蝠侠", "超人", "神奇女侠", "正义联盟", "闪电侠", "水行侠", "超级英雄", "漫画"],
+ "蝙蝠侠": ["布鲁斯韦恩", "高谭市", "小丑", "罗宾", "DC", "超级英雄"],
+ "超人": ["克拉克肯特", "氪星", "莱克斯卢瑟", "超能力", "DC", "超级英雄"],
+
+ "星球大战": ["星战", "原力", "天行者", "达斯维达", "尤达", "绝地武士", "光剑", "帝国", "科幻"],
+ "原力": ["绝地武士", "星球大战", "天行者", "尤达", "光剑", "西斯", "科幻"],
+
+ "哈利波特": ["魔法", "霍格沃茨", "魔杖", "魔法石", "伏地魔", "巫师", "奇幻", "魔幻"],
+ "魔法": ["巫师", "法术", "咒语", "哈利波特", "霍格沃茨", "魔杖", "奇幻", "魔幻"],
+
+ "科幻": ["未来", "太空", "星际", "外星人", "人工智能", "机器人", "时空", "星球大战", "星际穿越"],
+ "太空": ["宇宙", "星球", "卫星", "宇航员", "航天", "科幻", "星际", "外太空"],
+ "人工智能": ["AI", "机器学习", "深度学习", "神经网络", "机器人", "算法", "科技", "科幻"],
+
+ "动作": ["武打", "格斗", "功夫", "特技", "追逐", "冒险", "刺激", "爆破"],
+ "冒险": ["探险", "奇遇", "探索", "未知", "旅程", "冒险家", "刺激", "危险"],
+ "奇幻": ["魔法", "魔幻", "神话", "异世界", "精灵", "龙", "魔戒", "哈利波特"],
+
+ "悬疑": ["推理", "谜题", "侦探", "神秘", "悬念", "惊悚", "犯罪", "悬疑片"],
+ "推理": ["侦探", "线索", "谜题", "破案", "悬疑", "逻辑", "智力", "悬疑片"],
+
+ "恐怖": ["惊悚", "鬼怪", "恶魔", "惊吓", "血腥", "恐怖片", "心理恐惧", "超自然"],
+ "鬼怪": ["幽灵", "鬼魂", "妖怪", "超自然", "恐怖", "惊悚", "诡异", "恐怖片"],
+
+ "喜剧": ["搞笑", "幽默", "欢乐", "笑声", "喜剧片", "滑稽", "逗乐", "喜剧演员"],
+ "搞笑": ["幽默", "笑话", "喜剧", "逗乐", "滑稽", "欢乐", "喜剧片", "喜剧演员"],
+
+ "战争": ["军事", "战场", "士兵", "军队", "战役", "武器", "战争片", "历史战争"],
+ "军事": ["军队", "武器", "战争", "军人", "战略", "战术", "国防", "军事片"],
+
+ "剧情": ["情节", "故事", "叙事", "人物", "感人", "真实", "戏剧性", "剧情片"],
+ "历史": ["古代", "历史事件", "历史人物", "朝代", "文明", "历史片", "传记", "纪实"],
+
+ "纪录片": ["真实记录", "纪实", "历史", "自然", "科学", "社会", "文化", "探索"],
+ "动画": ["卡通", "动漫", "动画片", "动画电影", "CG", "3D动画", "手绘", "二次元"],
+
+ "音乐": ["歌曲", "旋律", "节奏", "乐器", "演唱", "音乐家", "音乐剧", "音乐会"],
+ "歌曲": ["歌词", "唱歌", "歌手", "流行歌曲", "音乐", "专辑", "单曲", "MV"],
+
+ "爱情": ["恋爱", "浪漫", "情侣", "爱情故事", "爱情片", "感情", "爱意", "约会"],
+ "浪漫": ["爱情", "情感", "温馨", "甜蜜", "爱意", "爱情片", "情侣", "表白"],
+
+ "Netflix": ["网飞", "流媒体", "自制剧", "电视剧", "纸牌屋", "怪奇物语", "王冠", "订阅"],
+ "迪士尼": ["米老鼠", "唐老鸭", "公主", "动画", "迪士尼乐园", "皮克斯", "童话", "漫威"],
+
+ "游戏": ["电子游戏", "游戏机", "主机游戏", "PC游戏", "手游", "网游", "单机", "多人游戏"],
+ "动漫": ["日本动画", "漫画", "二次元", "动画", "动画片", "ACGN", "宅文化", "御宅族"],
+
+ "日本": ["东京", "京都", "大阪", "日本文化", "日本料理", "樱花", "动漫", "武士道"],
+ "美国": ["纽约", "洛杉矶", "华盛顿", "美国文化", "好莱坞", "自由女神像", "美式"],
+
+ "教育": ["学习", "知识", "课程", "教学", "学校", "教科书", "老师", "学生"],
+ "技术": ["科技", "工程", "编程", "软件", "硬件", "开发", "技术革新", "IT"],
+
+ "监狱": ["越狱", "囚犯", "牢房", "服刑", "狱警"],
+ "越狱": ["监狱", "囚犯", "逃狱", "越狱计划", "监狱逃脱"]
+}
diff --git a/JWLLL/word2vec_helper.py b/JWLLL/word2vec_helper.py
new file mode 100644
index 0000000..ecd1a72
--- /dev/null
+++ b/JWLLL/word2vec_helper.py
@@ -0,0 +1,279 @@
+# word2vec_helper.py
+# Word2Vec模型加载与使用的辅助模块
+
+import os
+import numpy as np
+from gensim.models import KeyedVectors, Word2Vec
+import jieba
+import logging
+import time
+
+# 设置日志
+logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.INFO)
+
+class Word2VecHelper:
+ def __init__(self, model_path=None):
+ """
+ 初始化Word2Vec辅助类
+
+ 参数:
+ model_path: 预训练模型路径，支持word2vec格式和二进制格式
+ 如果为None，将使用默认路径或尝试下载小型模型
+ """
+ self.model = None
+
+ # 更改默认模型路径和备用选项
+ if model_path:
+ self.model_path = model_path
+ else:
+ # 首选路径 - 大型腾讯模型
+ primary_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
+ "models", "chinese_word2vec.bin")
+
+ # 备用路径 - 小型模型
+ backup_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
+ "models", "chinese_word2vec_small.bin")
+
+ if os.path.exists(primary_path):
+ self.model_path = primary_path
+ elif os.path.exists(backup_path):
+ self.model_path = backup_path
+ else:
+ # 如果都不存在，可以尝试自动下载小模型
+ self.model_path = primary_path
+ self._try_download_small_model()
+
+ self.initialized = False
+ # 缓存查询结果，提高性能
+ self.similarity_cache = {}
+ self.similar_words_cache = {}
+
+ def _try_download_small_model(self):
+ """尝试下载小型词向量模型作为备用选项"""
+ try:
+ import gensim.downloader as api
+ logging.info("尝试下载小型中文词向量模型...")
+
+ # 创建模型目录
+ os.makedirs(os.path.dirname(self.model_path), exist_ok=True)
+
+ # 尝试下载fastText的小型中文模型
+ small_model = api.load("fasttext-wiki-news-subwords-300")
+ small_model.save(self.model_path.replace(".bin", "_small.bin"))
+ logging.info(f"小型模型已下载并保存到 {self.model_path}")
+ except Exception as e:
+ logging.error(f"无法下载备用模型: {e}")
+
+ def load_model(self):
+ """加载Word2Vec模型"""
+ try:
+ start_time = time.time()
+ logging.info(f"开始加载Word2Vec模型: {self.model_path}")
+
+ # 判断文件扩展名，选择合适的加载方式
+ if self.model_path.endswith('.bin'):
+ # 加载二进制格式的模型
+ self.model = KeyedVectors.load_word2vec_format(self.model_path, binary=True)
+ else:
+ # 加载文本格式的模型或gensim模型
+ self.model = Word2Vec.load(self.model_path).wv
+
+ self.initialized = True
+ logging.info(f"Word2Vec模型加载完成，耗时 {time.time() - start_time:.2f} 秒")
+ logging.info(f"词向量维度: {self.model.vector_size}")
+ logging.info(f"词汇表大小: {len(self.model.index_to_key)}")
+ return True
+ except Exception as e:
+ logging.error(f"加载Word2Vec模型失败: {e}")
+ self.initialized = False
+ return False
+
+ def ensure_initialized(self):
+ """确保模型已初始化"""
+ if not self.initialized:
+ return self.load_model()
+ return True
+
+ def get_similar_words(self, word, topn=10, min_similarity=0.5):
+ """
+ 获取与给定词语最相似的词语列表
+
+ 参数:
+ word: 输入词语
+ topn: 返回相似词的数量
+ min_similarity: 最小相似度阈值
+ 返回:
+ 相似词列表，如果词不存在或模型未加载则返回空列表
+ """
+ if not self.ensure_initialized():
+ return []
+
+ # 检查缓存
+ cache_key = f"{word}_{topn}_{min_similarity}"
+ if cache_key in self.similar_words_cache:
+ return self.similar_words_cache[cache_key]
+
+ try:
+ # 如果词不在词汇表中，进行分词处理
+ if word not in self.model.key_to_index:
+ # 对中文词进行分词，然后查找每个子词的相似词
+ word_parts = list(jieba.cut(word))
+
+ if not word_parts:
+ return []
+
+ # 如果存在多个子词，找到存在于模型中的子词
+ valid_parts = [w for w in word_parts if w in self.model.key_to_index]
+
+ if not valid_parts:
+ return []
+
+ # 使用最长的有效子词或第一个有效子词
+ valid_parts.sort(key=len, reverse=True)
+ word = valid_parts[0]
+
+ # 如果替换后的词仍不在词汇表中，返回空列表
+ if word not in self.model.key_to_index:
+ return []
+
+ # 获取相似词
+ similar_words = self.model.most_similar(word, topn=topn*2) # 多获取一些，后续过滤
+
+ # 过滤低于阈值的结果，并只返回词语(不返回相似度)
+ filtered_words = [w for w, sim in similar_words if sim >= min_similarity][:topn]
+
+ # 缓存结果
+ self.similar_words_cache[cache_key] = filtered_words
+ return filtered_words
+
+ except Exception as e:
+ logging.error(f"获取相似词失败: {e}, 词语: {word}")
+ return []
+
+ def calculate_similarity(self, word1, word2):
+ """
+ 计算两个词的相似度
+
+ 参数:
+ word1, word2: 输入词语
+ 返回:
+ 相似度分数(0-1)，如果任意词不存在则返回0
+ """
+ if not self.ensure_initialized():
+ return 0
+
+ # 检查缓存
+ cache_key = f"{word1}_{word2}"
+ reverse_key = f"{word2}_{word1}"
+
+ if cache_key in self.similarity_cache:
+ return self.similarity_cache[cache_key]
+ if reverse_key in self.similarity_cache:
+ return self.similarity_cache[reverse_key]
+
+ try:
+ # 检查词是否在词汇表中
+ if word1 not in self.model.key_to_index or word2 not in self.model.key_to_index:
+ return 0
+
+ similarity = self.model.similarity(word1, word2)
+
+ # 缓存结果
+ self.similarity_cache[cache_key] = similarity
+ return similarity
+
+ except Exception as e:
+ logging.error(f"计算相似度失败: {e}, 词语: {word1}, {word2}")
+ return 0
+
+ def expand_query(self, query, topn=5, min_similarity=0.6):
+ """
+ 扩展查询词，返回相关词汇
+
+ 参数:
+ query: 查询词
+ topn: 每个词扩展的相似词数量
+ min_similarity: 最小相似度阈值
+ 返回:
+ 扩展后的词语列表
+ """
+ if not self.ensure_initialized():
+ return [query]
+
+ expanded_terms = [query]
+
+ # 对查询进行分词
+ words = list(jieba.cut(query))
+
+ # 为每个词找相似词
+ for word in words:
+ if len(word) <= 1: # 忽略单字，减少噪音
+ continue
+
+ similar_words = self.get_similar_words(word, topn=topn, min_similarity=min_similarity)
+ expanded_terms.extend(similar_words)
+
+ # 确保唯一性
+ return list(set(expanded_terms))
+
+# 单例模式，全局使用一个模型实例
+_word2vec_helper = None
+
+def get_word2vec_helper(model_path=None):
+ """获取Word2Vec辅助类的全局单例"""
+ global _word2vec_helper
+ if _word2vec_helper is None:
+ _word2vec_helper = Word2VecHelper(model_path)
+ _word2vec_helper.ensure_initialized()
+ return _word2vec_helper
+
+# 便捷函数，方便直接调用
+def get_similar_words(word, topn=10, min_similarity=0.5):
+ """获取相似词的便捷函数"""
+ helper = get_word2vec_helper()
+ return helper.get_similar_words(word, topn, min_similarity)
+
+def calculate_similarity(word1, word2):
+ """计算相似度的便捷函数"""
+ helper = get_word2vec_helper()
+ return helper.calculate_similarity(word1, word2)
+
+def expand_query(query, topn=5, min_similarity=0.6):
+ """扩展查询的便捷函数"""
+ helper = get_word2vec_helper()
+ return helper.expand_query(query, topn, min_similarity)
+
+# 使用示例
+if __name__ == "__main__":
+ # 测试模型加载和词语相似度
+ helper = get_word2vec_helper()
+
+ # 测试词
+ test_words = ["电影", "功夫", "熊猫", "科幻", "漫威"]
+
+ for word in test_words:
+ print(f"\n{word} 的相似词:")
+ similar = helper.get_similar_words(word, topn=5)
+ for sim_word in similar:
+ print(f" - {sim_word}")
+
+ # 测试相似度计算
+ word_pairs = [
+ ("电影", "电视"),
+ ("功夫", "武术"),
+ ("科幻", "未来"),
+ ("漫威", "超级英雄")
+ ]
+
+ print("\n词语相似度:")
+ for w1, w2 in word_pairs:
+ sim = helper.calculate_similarity(w1, w2)
+ print(f" {w1} <-> {w2}: {sim:.4f}")
+
+ # 测试查询扩展
+ test_queries = ["功夫熊猫", "科幻电影", "漫威英雄"]
+
+ print("\n查询扩展:")
+ for query in test_queries:
+ expanded = helper.expand_query(query)
+ print(f" {query} -> {expanded}")
diff --git "a/JWLLL/\346\216\245\345\217\243\346\265\213\350\257\225\350\257\264\346\230\216.md" "b/JWLLL/\346\216\245\345\217\243\346\265\213\350\257\225\350\257\264\346\230\216.md"
new file mode 100644
index 0000000..1537055
--- /dev/null
+++ "b/JWLLL/\346\216\245\345\217\243\346\265\213\350\257\225\350\257\264\346\230\216.md"
@@ -0,0 +1,186 @@
+# 接口说明文档
+
+本服务为资源搜索与推荐API，所有接口均支持Postman测试。每个接口均包含功能说明、请求方式、参数、返回值、详细Postman测试方法，并补充了核心逻辑和原理说明。
+
+---
+
+## 1. 搜索接口
+- **接口功能**：根据关键词、分类、标签等条件搜索资源。
+- **核心逻辑与原理**：
+ - 支持关键词分词、拼音、语义扩展（包括自定义语义映射和Word2Vec相似词扩展）。
+ - 支持多字段（标题、内容、分类、标签）模糊匹配。
+ - 相关性打分综合考虑精确匹配、分词、拼音、标签、描述、分类等多种因素。
+ - 支持多种排序方式（热度、时间、相似度等）。
+- **请求方式**：POST
+- **URL**：`/search`
+- **请求参数**：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | ----------- | ------- | ---- | -------------------------------------- |
+ | keyword | string | 是 | 搜索关键词 |
+ | sort_by | string | 否 | 排序方式（downloads、similarity等） |
+ | category | string | 否 | 分类名 |
+ | search_mode | string | 否 | 搜索模式（title、title_desc、all等） |
+ | tags | array | 否 | 标签数组 |
+- **返回说明**：
+ - results: 资源列表，每项包含id、title、category、heat、created_at等字段。
+
+**Postman测试方法：**
+1. 新建POST请求，URL填`http://127.0.0.1:5000/search`
+2. Body选择raw，类型JSON，内容示例：
+ ```json
+ {
+ "keyword": "电影",
+ "sort_by": "downloads"
+ }
+ ```
+3. 点击Send，查看返回结果。
+
+---
+
+## 2. 标签推荐接口
+- **接口功能**：根据用户兴趣标签推荐相关资源。
+- **核心逻辑与原理**：
+ - 首先根据用户兴趣标签（user_tags表+tags表）查找相关资源。
+ - 若无结果，则用标签名模糊匹配资源标题和内容。
+ - 推荐结果按相关性和热度排序。
+- **请求方式**：POST
+- **URL**：`/recommend_tags`
+- **请求参数**：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | -------- | ------- | ---- | -------------- |
+ | user_id | string | 是 | 用户ID |
+ | tags | array | 否 | 用户关注标签 |
+- **返回说明**：
+ - recommendations: 推荐资源列表。
+
+**Postman测试方法：**
+1. 新建POST请求，URL填`http://127.0.0.1:5000/recommend_tags`
+2. Body选择raw，类型JSON，内容示例：
+ ```json
+ {
+ "user_id": "1",
+ "tags": ["动作", "科幻"]
+ }
+ ```
+3. 点击Send，查看推荐结果。
+
+---
+
+## 3. 用户兴趣标签管理接口
+- **接口功能**：管理用户兴趣标签（增删查）。
+- **核心逻辑与原理**：
+ - 用户标签数据存储在 user_tags 表，通过 tag_id 关联 tags 表，所有操作均以标签名为主。
+ - 支持添加、删除、查询用户兴趣标签。
+- **请求方式**：POST/GET/DELETE
+- **URL**：`/user_tags` 或 `/tags`
+- **请求参数**：
+ - POST/DELETE：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | ------- | ------- | ---- | -------- |
+ | user_id | string | 是 | 用户ID |
+ | tags | array | 是 | 标签数组 |
+ - GET：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | ------- | ------- | ---- | -------- |
+ | user_id | string | 是 | 用户ID |
+- **返回说明**：
+ - tags: 用户当前兴趣标签列表。
+
+**Postman测试方法：**
+- 添加标签（POST）：
+ 1. 新建POST请求，URL填`http://127.0.0.1:5000/user_tags`
+ 2. Body选择raw，类型JSON：
+ ```json
+ {
+ "user_id": "1",
+ "tags": ["动作", "科幻"]
+ }
+ ```
+ 3. Send。
+- 查询标签（GET）：
+ 1. 新建GET请求，URL填`http://127.0.0.1:5000/user_tags?user_id=1`
+ 2. Send。
+- 删除标签（DELETE）：
+ 1. 新建DELETE请求，URL填`http://127.0.0.1:5000/user_tags`
+ 2. Body选择raw，类型JSON：
+ ```json
+ {
+ "user_id": "1",
+ "tags": ["动作", "科幻"]
+ }
+ ```
+ 3. Send。
+
+---
+
+## 4. 协同过滤推荐接口
+- **接口功能**：基于用户行为的个性化推荐。
+- **核心逻辑与原理**：
+ - 基于 behaviors 表的用户行为（type='favorite' 或 'view'）构建用户-物品矩阵。
+ - 计算用户与其他用户的兴趣重叠度（Jaccard相似度=交集/并集），找出最相似的用户。
+ - 推荐这些相似用户收藏/浏览过、但当前用户未看过的帖子。
+ - 若用户行为数据不足或无相似用户，则推荐全站热门资源。
+- **请求方式**：POST
+- **URL**：`/user_based_recommend`
+- **请求参数**：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | ------- | ------- | ---- | ------------ |
+ | user_id | string | 是 | 用户ID |
+ | top_n | int | 否 | 推荐数量 |
+- **返回说明**：
+ - recommendations: 推荐资源列表。
+
+**Postman测试方法：**
+1. 新建POST请求，URL填`http://127.0.0.1:5000/user_based_recommend`
+2. Body选择raw，类型JSON：
+ ```json
+ {
+ "user_id": "1",
+ "top_n": 3
+ }
+ ```
+3. Send。
+
+---
+
+## 5. Word2Vec状态检查接口
+- **接口功能**：检查Word2Vec模型加载状态。
+- **核心逻辑与原理**：
+ - 检查Word2Vec模型是否加载成功，返回词汇量、向量维度、部分词的相似词等信息。
+- **请求方式**：GET
+- **URL**：`/word2vec_status`
+- **返回说明**：
+ - enabled, initialized, vocab_size, vector_size, test_results等。
+
+**Postman测试方法：**
+1. 新建GET请求，URL填`http://127.0.0.1:5000/word2vec_status`
+2. Send。
+
+---
+
+## 6. 调试接口
+- **接口功能**：数据库调试与数据检查。
+- **核心逻辑与原理**：
+ - 通过多种SQL查询，辅助开发者调试数据库内容和搜索命中情况。
+- **请求方式**：POST
+- **URL**：`/debug_search`
+- **请求参数**：
+ | 参数名 | 类型 | 必填 | 说明 |
+ | ------- | ------- | ---- | -------- |
+ | keyword | string | 是 | 关键词 |
+- **返回说明**：
+ - 各类调试用的数据库查询结果。
+
+**Postman测试方法：**
+1. 新建POST请求，URL填`http://127.0.0.1:5000/debug_search`
+2. Body选择raw，类型JSON：
+ ```json
+ {
+ "keyword": "电影"
+ }
+ ```
+3. Send。
+
+---
+
+如需补充其它接口或参数说明，随时联系！